List of usage examples for org.apache.hadoop.yarn.client.api AMRMClient start
@Override public void start()
From source file:com.datatorrent.stram.StramMiniClusterTest.java
License:Apache License
@Ignore @Test//from w w w.j a v a 2s . co m public void testUnmanagedAM2() throws Exception { new InlineAM(conf) { @Override @SuppressWarnings("SleepWhileInLoop") public void runAM(ApplicationAttemptId attemptId) throws Exception { LOG.debug("AM running {}", attemptId); AMRMClient<ContainerRequest> amRmClient = AMRMClient.createAMRMClient(); amRmClient.init(conf); amRmClient.start(); // register with the RM (LAUNCHED -> RUNNING) amRmClient.registerApplicationMaster("", 0, null); // AM specific logic String[] hosts = { "vm1" }; String[] racks = { "somerack" }; Resource capability = Records.newRecord(Resource.class); capability.setMemory(1000); Priority priority = Records.newRecord(Priority.class); priority.setPriority(10); AMRMClient.ContainerRequest req = new AMRMClient.ContainerRequest(capability, hosts, racks, priority); amRmClient.addContainerRequest(req); amRmClient.addContainerRequest(req); /* capability = Records.newRecord(Resource.class); capability.setMemory(5512); priority = Records.newRecord(Priority.class); priority.setPriority(11); req = new AMRMClient.ContainerRequest(capability, hosts, racks, priority, 3); amRmClient.addContainerRequest(req); */ for (int i = 0; i < 100; i++) { AllocateResponse ar = amRmClient.allocate(0); sleep(1000); LOG.debug("allocateResponse: {}", ar); for (Container c : ar.getAllocatedContainers()) { LOG.debug("*** allocated {}", c.getResource()); amRmClient.removeContainerRequest(req); } /* GetClusterNodesRequest request = Records.newRecord(GetClusterNodesRequest.class); ClientRMService clientRMService = yarnCluster.getResourceManager().getClientRMService(); GetClusterNodesResponse response = clientRMService.getClusterNodes(request); List<NodeReport> nodeReports = response.getNodeReports(); LOG.info(nodeReports); for (NodeReport nr: nodeReports) { LOG.info("Node: " + nr.getNodeId()); LOG.info("Total memory: " + nr.getCapability()); LOG.info("Used memory: " + nr.getUsed()); LOG.info("Number containers: " + nr.getNumContainers()); } */ } // unregister from RM amRmClient.unregisterApplicationMaster(FinalApplicationStatus.SUCCEEDED, "testUnmanagedAM finished", null); } }.run(); }
From source file:com.ibm.bi.dml.yarn.DMLAppMaster.java
License:Open Source License
/** * //from w w w .j a va 2 s . co m * @param args * @throws YarnException * @throws IOException */ public void runApplicationMaster(String[] args) throws YarnException, IOException { _conf = new YarnConfiguration(); //obtain application ID String containerIdString = System.getenv(Environment.CONTAINER_ID.name()); ContainerId containerId = ConverterUtils.toContainerId(containerIdString); _appId = containerId.getApplicationAttemptId().getApplicationId(); LOG.info("SystemML appplication master (applicationID: " + _appId + ")"); //initialize clients to ResourceManager AMRMClient<ContainerRequest> rmClient = AMRMClient.createAMRMClient(); rmClient.init(_conf); rmClient.start(); //register with ResourceManager rmClient.registerApplicationMaster("", 0, ""); //host, port for rm communication LOG.debug("Registered the SystemML application master with resource manager"); //start status reporter to ResourceManager DMLAppMasterStatusReporter reporter = new DMLAppMasterStatusReporter(rmClient, 10000); reporter.start(); LOG.debug("Started status reporter (heartbeat to resource manager)"); //set DMLscript app master context DMLScript.setActiveAM(); //parse input arguments String[] otherArgs = new GenericOptionsParser(_conf, args).getRemainingArgs(); //run SystemML CP FinalApplicationStatus status = null; try { //core dml script execution (equivalent to non-AM runtime) boolean success = DMLScript.executeScript(_conf, otherArgs); if (success) status = FinalApplicationStatus.SUCCEEDED; else status = FinalApplicationStatus.FAILED; } catch (DMLScriptException ex) { LOG.error(DMLYarnClient.APPMASTER_NAME + ": Failed to executed DML script due to stop call:\n\t" + ex.getMessage()); status = FinalApplicationStatus.FAILED; writeMessageToHDFSWorkingDir(ex.getMessage()); } catch (Exception ex) { LOG.error(DMLYarnClient.APPMASTER_NAME + ": Failed to executed DML script.", ex); status = FinalApplicationStatus.FAILED; } finally { //stop periodic status reports reporter.stopStatusReporter(); LOG.debug("Stopped status reporter"); //unregister resource manager client rmClient.unregisterApplicationMaster(status, "", ""); LOG.debug("Unregistered the SystemML application master"); } }
From source file:eu.stratosphere.yarn.ApplicationMaster.java
License:Apache License
private void run() throws Exception { //Utils.logFilesInCurrentDirectory(LOG); // Initialize clients to ResourceManager and NodeManagers Configuration conf = Utils.initializeYarnConfiguration(); FileSystem fs = FileSystem.get(conf); Map<String, String> envs = System.getenv(); final String currDir = envs.get(Environment.PWD.key()); final String logDirs = envs.get(Environment.LOG_DIRS.key()); final String ownHostname = envs.get(Environment.NM_HOST.key()); final String appId = envs.get(Client.ENV_APP_ID); final String clientHomeDir = envs.get(Client.ENV_CLIENT_HOME_DIR); final String applicationMasterHost = envs.get(Environment.NM_HOST.key()); final String remoteStratosphereJarPath = envs.get(Client.STRATOSPHERE_JAR_PATH); final String shipListString = envs.get(Client.ENV_CLIENT_SHIP_FILES); final String yarnClientUsername = envs.get(Client.ENV_CLIENT_USERNAME); final int taskManagerCount = Integer.valueOf(envs.get(Client.ENV_TM_COUNT)); final int memoryPerTaskManager = Integer.valueOf(envs.get(Client.ENV_TM_MEMORY)); final int coresPerTaskManager = Integer.valueOf(envs.get(Client.ENV_TM_CORES)); int heapLimit = Utils.calculateHeapSize(memoryPerTaskManager); if (currDir == null) { throw new RuntimeException("Current directory unknown"); }/*from www . j av a2s . c o m*/ if (ownHostname == null) { throw new RuntimeException("Own hostname (" + Environment.NM_HOST + ") not set."); } LOG.info("Working directory " + currDir); // load Stratosphere configuration. Utils.getStratosphereConfiguration(currDir); final String localWebInterfaceDir = currDir + "/resources/" + ConfigConstants.DEFAULT_JOB_MANAGER_WEB_PATH_NAME; // Update yaml conf -> set jobManager address to this machine's address. FileInputStream fis = new FileInputStream(currDir + "/stratosphere-conf.yaml"); BufferedReader br = new BufferedReader(new InputStreamReader(fis)); Writer output = new BufferedWriter(new FileWriter(currDir + "/stratosphere-conf-modified.yaml")); String line; while ((line = br.readLine()) != null) { if (line.contains(ConfigConstants.JOB_MANAGER_IPC_ADDRESS_KEY)) { output.append(ConfigConstants.JOB_MANAGER_IPC_ADDRESS_KEY + ": " + ownHostname + "\n"); } else if (line.contains(ConfigConstants.JOB_MANAGER_WEB_ROOT_PATH_KEY)) { output.append(ConfigConstants.JOB_MANAGER_WEB_ROOT_PATH_KEY + ": " + "\n"); } else { output.append(line + "\n"); } } // just to make sure. output.append(ConfigConstants.JOB_MANAGER_IPC_ADDRESS_KEY + ": " + ownHostname + "\n"); output.append(ConfigConstants.JOB_MANAGER_WEB_ROOT_PATH_KEY + ": " + localWebInterfaceDir + "\n"); output.append(ConfigConstants.JOB_MANAGER_WEB_LOG_PATH_KEY + ": " + logDirs + "\n"); output.close(); br.close(); File newConf = new File(currDir + "/stratosphere-conf-modified.yaml"); if (!newConf.exists()) { LOG.warn("modified yaml does not exist!"); } Utils.copyJarContents("resources/" + ConfigConstants.DEFAULT_JOB_MANAGER_WEB_PATH_NAME, ApplicationMaster.class.getProtectionDomain().getCodeSource().getLocation().getPath()); JobManager jm; { String pathToNepheleConfig = currDir + "/stratosphere-conf-modified.yaml"; String[] args = { "-executionMode", "cluster", "-configDir", pathToNepheleConfig }; // start the job manager jm = JobManager.initialize(args); // Start info server for jobmanager jm.startInfoServer(); } AMRMClient<ContainerRequest> rmClient = AMRMClient.createAMRMClient(); rmClient.init(conf); rmClient.start(); NMClient nmClient = NMClient.createNMClient(); nmClient.init(conf); nmClient.start(); // Register with ResourceManager LOG.info("registering ApplicationMaster"); rmClient.registerApplicationMaster(applicationMasterHost, 0, "http://" + applicationMasterHost + ":" + GlobalConfiguration.getString(ConfigConstants.JOB_MANAGER_WEB_PORT_KEY, "undefined")); // Priority for worker containers - priorities are intra-application Priority priority = Records.newRecord(Priority.class); priority.setPriority(0); // Resource requirements for worker containers Resource capability = Records.newRecord(Resource.class); capability.setMemory(memoryPerTaskManager); capability.setVirtualCores(coresPerTaskManager); // Make container requests to ResourceManager for (int i = 0; i < taskManagerCount; ++i) { ContainerRequest containerAsk = new ContainerRequest(capability, null, null, priority); LOG.info("Requesting TaskManager container " + i); rmClient.addContainerRequest(containerAsk); } LocalResource stratosphereJar = Records.newRecord(LocalResource.class); LocalResource stratosphereConf = Records.newRecord(LocalResource.class); // register Stratosphere Jar with remote HDFS final Path remoteJarPath = new Path(remoteStratosphereJarPath); Utils.registerLocalResource(fs, remoteJarPath, stratosphereJar); // register conf with local fs. Path remoteConfPath = Utils.setupLocalResource(conf, fs, appId, new Path("file://" + currDir + "/stratosphere-conf-modified.yaml"), stratosphereConf, new Path(clientHomeDir)); LOG.info("Prepared localresource for modified yaml: " + stratosphereConf); boolean hasLog4j = new File(currDir + "/log4j.properties").exists(); // prepare the files to ship LocalResource[] remoteShipRsc = null; String[] remoteShipPaths = shipListString.split(","); if (!shipListString.isEmpty()) { remoteShipRsc = new LocalResource[remoteShipPaths.length]; { // scope for i int i = 0; for (String remoteShipPathStr : remoteShipPaths) { if (remoteShipPathStr == null || remoteShipPathStr.isEmpty()) { continue; } remoteShipRsc[i] = Records.newRecord(LocalResource.class); Path remoteShipPath = new Path(remoteShipPathStr); Utils.registerLocalResource(fs, remoteShipPath, remoteShipRsc[i]); i++; } } } // respect custom JVM options in the YAML file final String javaOpts = GlobalConfiguration.getString(ConfigConstants.STRATOSPHERE_JVM_OPTIONS, ""); // Obtain allocated containers and launch int allocatedContainers = 0; int completedContainers = 0; while (allocatedContainers < taskManagerCount) { AllocateResponse response = rmClient.allocate(0); for (Container container : response.getAllocatedContainers()) { LOG.info("Got new Container for TM " + container.getId() + " on host " + container.getNodeId().getHost()); ++allocatedContainers; // Launch container by create ContainerLaunchContext ContainerLaunchContext ctx = Records.newRecord(ContainerLaunchContext.class); String tmCommand = "$JAVA_HOME/bin/java -Xmx" + heapLimit + "m " + javaOpts; if (hasLog4j) { tmCommand += " -Dlog.file=\"" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/taskmanager-log4j.log\" -Dlog4j.configuration=file:log4j.properties"; } tmCommand += " eu.stratosphere.yarn.YarnTaskManagerRunner -configDir . " + " 1>" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/taskmanager-stdout.log" + " 2>" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/taskmanager-stderr.log"; ctx.setCommands(Collections.singletonList(tmCommand)); LOG.info("Starting TM with command=" + tmCommand); // copy resources to the TaskManagers. Map<String, LocalResource> localResources = new HashMap<String, LocalResource>(2); localResources.put("stratosphere.jar", stratosphereJar); localResources.put("stratosphere-conf.yaml", stratosphereConf); // add ship resources if (!shipListString.isEmpty()) { Preconditions.checkNotNull(remoteShipRsc); for (int i = 0; i < remoteShipPaths.length; i++) { localResources.put(new Path(remoteShipPaths[i]).getName(), remoteShipRsc[i]); } } ctx.setLocalResources(localResources); // Setup CLASSPATH for Container (=TaskTracker) Map<String, String> containerEnv = new HashMap<String, String>(); Utils.setupEnv(conf, containerEnv); //add stratosphere.jar to class path. containerEnv.put(Client.ENV_CLIENT_USERNAME, yarnClientUsername); ctx.setEnvironment(containerEnv); UserGroupInformation user = UserGroupInformation.getCurrentUser(); try { Credentials credentials = user.getCredentials(); DataOutputBuffer dob = new DataOutputBuffer(); credentials.writeTokenStorageToStream(dob); ByteBuffer securityTokens = ByteBuffer.wrap(dob.getData(), 0, dob.getLength()); ctx.setTokens(securityTokens); } catch (IOException e) { LOG.warn("Getting current user info failed when trying to launch the container" + e.getMessage()); } LOG.info("Launching container " + allocatedContainers); nmClient.startContainer(container, ctx); } for (ContainerStatus status : response.getCompletedContainersStatuses()) { ++completedContainers; LOG.info("Completed container (while allocating) " + status.getContainerId() + ". Total Completed:" + completedContainers); LOG.info("Diagnostics " + status.getDiagnostics()); } Thread.sleep(100); } // Now wait for containers to complete while (completedContainers < taskManagerCount) { AllocateResponse response = rmClient.allocate(completedContainers / taskManagerCount); for (ContainerStatus status : response.getCompletedContainersStatuses()) { ++completedContainers; LOG.info("Completed container " + status.getContainerId() + ". Total Completed:" + completedContainers); LOG.info("Diagnostics " + status.getDiagnostics()); } Thread.sleep(5000); } LOG.info("Shutting down JobManager"); jm.shutdown(); // Un-register with ResourceManager rmClient.unregisterApplicationMaster(FinalApplicationStatus.SUCCEEDED, "", ""); }
From source file:org.apache.flink.yarn.ApplicationMaster.java
License:Apache License
private void run() throws Exception { //Utils.logFilesInCurrentDirectory(LOG); // Initialize clients to ResourceManager and NodeManagers Configuration conf = Utils.initializeYarnConfiguration(); FileSystem fs = FileSystem.get(conf); Map<String, String> envs = System.getenv(); final String currDir = envs.get(Environment.PWD.key()); final String logDirs = envs.get(Environment.LOG_DIRS.key()); final String ownHostname = envs.get(Environment.NM_HOST.key()); final String appId = envs.get(Client.ENV_APP_ID); final String clientHomeDir = envs.get(Client.ENV_CLIENT_HOME_DIR); final String applicationMasterHost = envs.get(Environment.NM_HOST.key()); final String remoteFlinkJarPath = envs.get(Client.FLINK_JAR_PATH); final String shipListString = envs.get(Client.ENV_CLIENT_SHIP_FILES); final String yarnClientUsername = envs.get(Client.ENV_CLIENT_USERNAME); final int taskManagerCount = Integer.valueOf(envs.get(Client.ENV_TM_COUNT)); final int memoryPerTaskManager = Integer.valueOf(envs.get(Client.ENV_TM_MEMORY)); final int coresPerTaskManager = Integer.valueOf(envs.get(Client.ENV_TM_CORES)); int heapLimit = Utils.calculateHeapSize(memoryPerTaskManager); if (currDir == null) { throw new RuntimeException("Current directory unknown"); }//from w ww .j a va 2 s.co m if (ownHostname == null) { throw new RuntimeException("Own hostname (" + Environment.NM_HOST + ") not set."); } LOG.info("Working directory " + currDir); // load Flink configuration. Utils.getFlinkConfiguration(currDir); final String localWebInterfaceDir = currDir + "/resources/" + ConfigConstants.DEFAULT_JOB_MANAGER_WEB_PATH_NAME; // Update yaml conf -> set jobManager address to this machine's address. FileInputStream fis = new FileInputStream(currDir + "/flink-conf.yaml"); BufferedReader br = new BufferedReader(new InputStreamReader(fis)); Writer output = new BufferedWriter(new FileWriter(currDir + "/flink-conf-modified.yaml")); String line; while ((line = br.readLine()) != null) { if (line.contains(ConfigConstants.JOB_MANAGER_IPC_ADDRESS_KEY)) { output.append(ConfigConstants.JOB_MANAGER_IPC_ADDRESS_KEY + ": " + ownHostname + "\n"); } else if (line.contains(ConfigConstants.JOB_MANAGER_WEB_ROOT_PATH_KEY)) { output.append(ConfigConstants.JOB_MANAGER_WEB_ROOT_PATH_KEY + ": " + "\n"); } else { output.append(line + "\n"); } } // just to make sure. output.append(ConfigConstants.JOB_MANAGER_IPC_ADDRESS_KEY + ": " + ownHostname + "\n"); output.append(ConfigConstants.JOB_MANAGER_WEB_ROOT_PATH_KEY + ": " + localWebInterfaceDir + "\n"); output.append(ConfigConstants.JOB_MANAGER_WEB_LOG_PATH_KEY + ": " + logDirs + "\n"); output.close(); br.close(); File newConf = new File(currDir + "/flink-conf-modified.yaml"); if (!newConf.exists()) { LOG.warn("modified yaml does not exist!"); } Utils.copyJarContents("resources/" + ConfigConstants.DEFAULT_JOB_MANAGER_WEB_PATH_NAME, ApplicationMaster.class.getProtectionDomain().getCodeSource().getLocation().getPath()); JobManager jm; { String pathToNepheleConfig = currDir + "/flink-conf-modified.yaml"; String[] args = { "-executionMode", "cluster", "-configDir", pathToNepheleConfig }; // start the job manager jm = JobManager.initialize(args); // Start info server for jobmanager jm.startInfoServer(); } AMRMClient<ContainerRequest> rmClient = AMRMClient.createAMRMClient(); rmClient.init(conf); rmClient.start(); NMClient nmClient = NMClient.createNMClient(); nmClient.init(conf); nmClient.start(); // Register with ResourceManager LOG.info("registering ApplicationMaster"); rmClient.registerApplicationMaster(applicationMasterHost, 0, "http://" + applicationMasterHost + ":" + GlobalConfiguration.getString(ConfigConstants.JOB_MANAGER_WEB_PORT_KEY, "undefined")); // Priority for worker containers - priorities are intra-application Priority priority = Records.newRecord(Priority.class); priority.setPriority(0); // Resource requirements for worker containers Resource capability = Records.newRecord(Resource.class); capability.setMemory(memoryPerTaskManager); capability.setVirtualCores(coresPerTaskManager); // Make container requests to ResourceManager for (int i = 0; i < taskManagerCount; ++i) { ContainerRequest containerAsk = new ContainerRequest(capability, null, null, priority); LOG.info("Requesting TaskManager container " + i); rmClient.addContainerRequest(containerAsk); } LocalResource flinkJar = Records.newRecord(LocalResource.class); LocalResource flinkConf = Records.newRecord(LocalResource.class); // register Flink Jar with remote HDFS final Path remoteJarPath = new Path(remoteFlinkJarPath); Utils.registerLocalResource(fs, remoteJarPath, flinkJar); // register conf with local fs. Path remoteConfPath = Utils.setupLocalResource(conf, fs, appId, new Path("file://" + currDir + "/flink-conf-modified.yaml"), flinkConf, new Path(clientHomeDir)); LOG.info("Prepared localresource for modified yaml: " + flinkConf); boolean hasLog4j = new File(currDir + "/log4j.properties").exists(); // prepare the files to ship LocalResource[] remoteShipRsc = null; String[] remoteShipPaths = shipListString.split(","); if (!shipListString.isEmpty()) { remoteShipRsc = new LocalResource[remoteShipPaths.length]; { // scope for i int i = 0; for (String remoteShipPathStr : remoteShipPaths) { if (remoteShipPathStr == null || remoteShipPathStr.isEmpty()) { continue; } remoteShipRsc[i] = Records.newRecord(LocalResource.class); Path remoteShipPath = new Path(remoteShipPathStr); Utils.registerLocalResource(fs, remoteShipPath, remoteShipRsc[i]); i++; } } } // respect custom JVM options in the YAML file final String javaOpts = GlobalConfiguration.getString(ConfigConstants.FLINK_JVM_OPTIONS, ""); // Obtain allocated containers and launch int allocatedContainers = 0; int completedContainers = 0; while (allocatedContainers < taskManagerCount) { AllocateResponse response = rmClient.allocate(0); for (Container container : response.getAllocatedContainers()) { LOG.info("Got new Container for TM " + container.getId() + " on host " + container.getNodeId().getHost()); ++allocatedContainers; // Launch container by create ContainerLaunchContext ContainerLaunchContext ctx = Records.newRecord(ContainerLaunchContext.class); String tmCommand = "$JAVA_HOME/bin/java -Xmx" + heapLimit + "m " + javaOpts; if (hasLog4j) { tmCommand += " -Dlog.file=\"" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/taskmanager-log4j.log\" -Dlog4j.configuration=file:log4j.properties"; } tmCommand += " org.apache.flink.yarn.YarnTaskManagerRunner -configDir . " + " 1>" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/taskmanager-stdout.log" + " 2>" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/taskmanager-stderr.log"; ctx.setCommands(Collections.singletonList(tmCommand)); LOG.info("Starting TM with command=" + tmCommand); // copy resources to the TaskManagers. Map<String, LocalResource> localResources = new HashMap<String, LocalResource>(2); localResources.put("flink.jar", flinkJar); localResources.put("flink-conf.yaml", flinkConf); // add ship resources if (!shipListString.isEmpty()) { Preconditions.checkNotNull(remoteShipRsc); for (int i = 0; i < remoteShipPaths.length; i++) { localResources.put(new Path(remoteShipPaths[i]).getName(), remoteShipRsc[i]); } } ctx.setLocalResources(localResources); // Setup CLASSPATH for Container (=TaskTracker) Map<String, String> containerEnv = new HashMap<String, String>(); Utils.setupEnv(conf, containerEnv); //add flink.jar to class path. containerEnv.put(Client.ENV_CLIENT_USERNAME, yarnClientUsername); ctx.setEnvironment(containerEnv); UserGroupInformation user = UserGroupInformation.getCurrentUser(); try { Credentials credentials = user.getCredentials(); DataOutputBuffer dob = new DataOutputBuffer(); credentials.writeTokenStorageToStream(dob); ByteBuffer securityTokens = ByteBuffer.wrap(dob.getData(), 0, dob.getLength()); ctx.setTokens(securityTokens); } catch (IOException e) { LOG.warn("Getting current user info failed when trying to launch the container" + e.getMessage()); } LOG.info("Launching container " + allocatedContainers); nmClient.startContainer(container, ctx); } for (ContainerStatus status : response.getCompletedContainersStatuses()) { ++completedContainers; LOG.info("Completed container (while allocating) " + status.getContainerId() + ". Total Completed:" + completedContainers); LOG.info("Diagnostics " + status.getDiagnostics()); } Thread.sleep(100); } // Now wait for containers to complete while (completedContainers < taskManagerCount) { AllocateResponse response = rmClient.allocate(completedContainers / taskManagerCount); for (ContainerStatus status : response.getCompletedContainersStatuses()) { ++completedContainers; LOG.info("Completed container " + status.getContainerId() + ". Total Completed:" + completedContainers); LOG.info("Diagnostics " + status.getDiagnostics()); } Thread.sleep(5000); } LOG.info("Shutting down JobManager"); jm.shutdown(); // Un-register with ResourceManager rmClient.unregisterApplicationMaster(FinalApplicationStatus.SUCCEEDED, "", ""); }
From source file:org.apache.flink.yarn.appMaster.ApplicationMaster.java
License:Apache License
public static void main(String[] args) throws Exception { // execute Application Master using the client's user final String yarnClientUsername = System.getenv(Client.ENV_CLIENT_USERNAME); LOG.info("YARN daemon runs as '" + UserGroupInformation.getCurrentUser().getShortUserName() + "' setting" + " user to execute Flink ApplicationMaster/JobManager to '" + yarnClientUsername + "'"); UserGroupInformation ugi = UserGroupInformation.createRemoteUser(yarnClientUsername); for (Token<? extends TokenIdentifier> toks : UserGroupInformation.getCurrentUser().getTokens()) { ugi.addToken(toks);/*from w w w .j a va2s . co m*/ } ugi.doAs(new PrivilegedAction<Object>() { @Override public Object run() { AMRMClient<ContainerRequest> rmClient = null; ApplicationMaster am = null; try { Configuration conf = Utils.initializeYarnConfiguration(); rmClient = AMRMClient.createAMRMClient(); rmClient.init(conf); rmClient.start(); // run the actual Application Master am = new ApplicationMaster(conf); am.generateConfigurationFile(); am.startJobManager(); am.setRMClient(rmClient); am.run(); } catch (Throwable e) { LOG.error("Error while running the application master", e); // the AM is not available. Report error through the unregister function. if (rmClient != null && am == null) { try { rmClient.unregisterApplicationMaster(FinalApplicationStatus.FAILED, "Flink YARN Application master" + " stopped unexpectedly with an exception.\n" + StringUtils.stringifyException(e), ""); } catch (Exception e1) { LOG.error("Unable to fail the application master", e1); } LOG.info("AM unregistered from RM"); return null; } if (rmClient == null) { LOG.error("Unable to unregister AM since the RM client is not available"); } if (am != null) { LOG.info("Writing error into internal message system"); am.setFailed(true); am.addMessage(new Message("The application master failed with an exception:\n" + StringUtils.stringifyException(e))); am.keepRPCAlive(); } } return null; } }); }
From source file:org.apache.sysml.yarn.DMLAppMaster.java
License:Apache License
public void runApplicationMaster(String[] args) throws YarnException, IOException { _conf = new YarnConfiguration(); //obtain application ID String containerIdString = System.getenv(Environment.CONTAINER_ID.name()); ContainerId containerId = ConverterUtils.toContainerId(containerIdString); _appId = containerId.getApplicationAttemptId().getApplicationId(); LOG.info("SystemML appplication master (applicationID: " + _appId + ")"); //initialize clients to ResourceManager AMRMClient<ContainerRequest> rmClient = AMRMClient.createAMRMClient(); rmClient.init(_conf);//from w w w . jav a 2 s. c o m rmClient.start(); //register with ResourceManager rmClient.registerApplicationMaster("", 0, ""); //host, port for rm communication LOG.debug("Registered the SystemML application master with resource manager"); //start status reporter to ResourceManager DMLAppMasterStatusReporter reporter = new DMLAppMasterStatusReporter(rmClient, 10000); reporter.start(); LOG.debug("Started status reporter (heartbeat to resource manager)"); //set DMLscript app master context DMLScript.setActiveAM(); //parse input arguments String[] otherArgs = new GenericOptionsParser(_conf, args).getRemainingArgs(); //run SystemML CP FinalApplicationStatus status = null; try { //core dml script execution (equivalent to non-AM runtime) boolean success = DMLScript.executeScript(_conf, otherArgs); if (success) status = FinalApplicationStatus.SUCCEEDED; else status = FinalApplicationStatus.FAILED; } catch (DMLScriptException ex) { LOG.error(DMLYarnClient.APPMASTER_NAME + ": Failed to executed DML script due to stop call:\n\t" + ex.getMessage()); status = FinalApplicationStatus.FAILED; writeMessageToHDFSWorkingDir(ex.getMessage()); } catch (Exception ex) { LOG.error(DMLYarnClient.APPMASTER_NAME + ": Failed to executed DML script.", ex); status = FinalApplicationStatus.FAILED; } finally { //stop periodic status reports reporter.stopStatusReporter(); LOG.debug("Stopped status reporter"); //unregister resource manager client rmClient.unregisterApplicationMaster(status, "", ""); LOG.debug("Unregistered the SystemML application master"); } }
From source file:org.chenchun.ApplicationMaster.java
License:Apache License
public static void main(String[] args) throws IOException, YarnException, InterruptedException { final String params = args[0]; final int containerNum = Integer.valueOf(args[1]); // Initialize clients to ResourceManager and NodeManagers Configuration conf = new YarnConfiguration(); AMRMClient<AMRMClient.ContainerRequest> rmClient = AMRMClient.createAMRMClient(); rmClient.init(conf);//from ww w. j a v a 2s. c o m rmClient.start(); NMClient nmClient = NMClient.createNMClient(); nmClient.init(conf); nmClient.start(); // Register with ResourceManager System.out.println("registerApplicationMaster 0"); rmClient.registerApplicationMaster("", 0, ""); System.out.println("registerApplicationMaster 1"); // Priority for worker containers - priorities are intra-application Priority priority = Records.newRecord(Priority.class); priority.setPriority(0); // Resource requirements for worker containers Resource capability = Records.newRecord(Resource.class); capability.setMemory(128); capability.setVirtualCores(1); // Make container requests to ResourceManager for (int i = 0; i < containerNum; ++i) { AMRMClient.ContainerRequest containerAsk = new AMRMClient.ContainerRequest(capability, null, null, priority); System.out.println("Making res-req " + i); rmClient.addContainerRequest(containerAsk); } // Obtain allocated containers, launch and check for responses int responseId = 0; int completedContainers = 0; while (completedContainers < containerNum) { AllocateResponse response = rmClient.allocate(responseId++); System.out.println("Allocate response " + response.getAMCommand() + " " + "allocate " + response.getAllocatedContainers().size() + "contains"); for (Container container : response.getAllocatedContainers()) { // Launch container by create ContainerLaunchContext ContainerLaunchContext ctx = Records.newRecord(ContainerLaunchContext.class); ctx.setCommands( Collections.singletonList(params + " 1>" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/stdout" + " 2>" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/stderr")); System.out.println("Launching container " + container.getId() + " on " + container.getNodeId()); nmClient.startContainer(container, ctx); } for (ContainerStatus status : response.getCompletedContainersStatuses()) { ++completedContainers; System.out.println("Completed container " + status.getContainerId()); } Thread.sleep(1000); } System.out.println("Unregister ApplicationMaster"); // Un-register with ResourceManager rmClient.unregisterApplicationMaster(FinalApplicationStatus.SUCCEEDED, "", ""); }
From source file:runtime.starter.MPJAppMaster.java
License:Open Source License
public void run() throws Exception { try {//from w w w.j a v a 2s.c o m appMasterSock = new Socket(serverName, ioServerPort); //redirecting stdout and stderr System.setOut(new PrintStream(appMasterSock.getOutputStream(), true)); System.setErr(new PrintStream(appMasterSock.getOutputStream(), true)); } catch (Exception exp) { exp.printStackTrace(); } FileSystem fs = FileSystem.get(conf); Path wrapperDest = new Path(wrapperPath); FileStatus destStatus = fs.getFileStatus(wrapperDest); Path userFileDest = new Path(userJarPath); FileStatus destStatusClass = fs.getFileStatus(userFileDest); // Initialize AM <--> RM communication protocol AMRMClient<ContainerRequest> rmClient = AMRMClient.createAMRMClient(); rmClient.init(conf); rmClient.start(); // Initialize AM <--> NM communication protocol NMClient nmClient = NMClient.createNMClient(); nmClient.init(conf); nmClient.start(); // Register with ResourceManager RegisterApplicationMasterResponse registerResponse = rmClient.registerApplicationMaster("", 0, ""); // Priority for containers - priorities are intra-application Priority priority = Records.newRecord(Priority.class); priority.setPriority(mpjContainerPriority); maxMem = registerResponse.getMaximumResourceCapability().getMemory(); if (debugYarn) { System.out.println("[MPJAppMaster]: Max memory capability resources " + "in cluster: " + maxMem); } if (containerMem > maxMem) { System.out.println("[MPJAppMaster]: container memory specified above " + "threshold of cluster! Using maximum memory for " + "containers: " + containerMem); containerMem = maxMem; } maxCores = registerResponse.getMaximumResourceCapability().getVirtualCores(); if (debugYarn) { System.out.println("[MPJAppMaster]: Max v-cores capability resources " + "in cluster: " + maxCores); } if (containerCores > maxCores) { System.out.println("[MPJAppMaster]: virtual cores specified above " + "threshold of cluster! Using maximum v-cores for " + "containers: " + containerCores); containerCores = maxCores; } // Resource requirements for containers Resource capability = Records.newRecord(Resource.class); capability.setMemory(containerMem); capability.setVirtualCores(containerCores); // Make container requests to ResourceManager for (int i = 0; i < np; ++i) { ContainerRequest containerReq = new ContainerRequest(capability, null, null, priority); rmClient.addContainerRequest(containerReq); } Map<String, LocalResource> localResources = new HashMap<String, LocalResource>(); // Creating Local Resource for Wrapper LocalResource wrapperJar = Records.newRecord(LocalResource.class); wrapperJar.setResource(ConverterUtils.getYarnUrlFromPath(wrapperDest)); wrapperJar.setSize(destStatus.getLen()); wrapperJar.setTimestamp(destStatus.getModificationTime()); wrapperJar.setType(LocalResourceType.ARCHIVE); wrapperJar.setVisibility(LocalResourceVisibility.APPLICATION); // Creating Local Resource for UserClass LocalResource userClass = Records.newRecord(LocalResource.class); userClass.setResource(ConverterUtils.getYarnUrlFromPath(userFileDest)); userClass.setSize(destStatusClass.getLen()); userClass.setTimestamp(destStatusClass.getModificationTime()); userClass.setType(LocalResourceType.ARCHIVE); userClass.setVisibility(LocalResourceVisibility.APPLICATION); localResources.put("mpj-yarn-wrapper.jar", wrapperJar); localResources.put("user-code.jar", userClass); while (allocatedContainers < np) { AllocateResponse response = rmClient.allocate(0); mpiContainers.addAll(response.getAllocatedContainers()); allocatedContainers = mpiContainers.size(); if (allocatedContainers != np) { Thread.sleep(100); } } if (debugYarn) { System.out.println("[MPJAppMaster]: launching " + allocatedContainers + " containers"); } for (Container container : mpiContainers) { ContainerLaunchContext ctx = Records.newRecord(ContainerLaunchContext.class); List<String> commands = new ArrayList<String>(); commands.add(" $JAVA_HOME/bin/java"); commands.add(" -Xmx" + containerMem + "m"); commands.add(" runtime.starter.MPJYarnWrapper"); commands.add("--serverName"); commands.add(serverName); // server name commands.add("--ioServerPort"); commands.add(Integer.toString(ioServerPort)); // IO server port commands.add("--deviceName"); commands.add(deviceName); // device name commands.add("--className"); commands.add(className); // class name commands.add("--psl"); commands.add(psl); // protocol switch limit commands.add("--np"); commands.add(Integer.toString(np)); // no. of containers commands.add("--rank"); commands.add(" " + Integer.toString(rank++)); // rank //temp sock port to share rank and ports commands.add("--wireUpPort"); commands.add(wireUpPort); if (appArgs != null) { commands.add("--appArgs"); for (int i = 0; i < appArgs.length; i++) { commands.add(appArgs[i]); } } ctx.setCommands(commands); // Set local resource for containers ctx.setLocalResources(localResources); // Set environment for container Map<String, String> containerEnv = new HashMap<String, String>(); setupEnv(containerEnv); ctx.setEnvironment(containerEnv); // Time to start the container nmClient.startContainer(container, ctx); } while (completedContainers < np) { // argument to allocate() is the progress indicator AllocateResponse response = rmClient.allocate(completedContainers / np); for (ContainerStatus status : response.getCompletedContainersStatuses()) { if (debugYarn) { System.out.println("\n[MPJAppMaster]: Container Id - " + status.getContainerId()); System.out.println("[MPJAppMaster]: Container State - " + status.getState().toString()); System.out.println("[MPJAppMaster]: Container Diagnostics - " + status.getDiagnostics()); } ++completedContainers; } if (completedContainers != np) { Thread.sleep(100); } ; } // Un-register with ResourceManager rmClient.unregisterApplicationMaster(FinalApplicationStatus.SUCCEEDED, "", ""); //shutDown AppMaster IO System.out.println("EXIT"); }
From source file:yrun.YarnRunnerApplicationMaster.java
License:Apache License
public void run(String[] args) throws Exception { JsonParser parser = new JsonParser(); JsonElement element = parser.parse(new FileReader(YarnRunner.MASTER_JSON)); LOG.info("Json [" + element + "]"); JsonObject jsonObject = (JsonObject) element; int priority = jsonObject.get("priority").getAsInt(); int numberOfContainers = jsonObject.get("numberOfContainers").getAsInt(); int memory = jsonObject.get("memory").getAsInt(); int vCores = jsonObject.get("vCores").getAsInt(); String command = jsonObject.get("command").getAsString(); // startHttpServer(); // InetSocketAddress address = _server.getAddress(); // LOG.info("Http server started at [" + address + "]"); // String appHostName = "app-host-name"; // int appHostPort = address.getPort(); // String appTrackingUrl = "http://" + address.getHostName() + ":" + // appHostPort + "/"; // LOG.info("App Tracking Url [" + appTrackingUrl + "]"); // Initialize clients to ResourceManager and NodeManagers Configuration conf = new YarnConfiguration(); AMRMClient<ContainerRequest> rmClient = AMRMClient.createAMRMClient(); rmClient.init(conf);//from w w w. jav a 2 s. c o m rmClient.start(); NMClient nmClient = NMClient.createNMClient(); nmClient.init(conf); nmClient.start(); // Register with ResourceManager LOG.info("Register Application Master 0"); String appHostName = ""; int appHostPort = 0; String appTrackingUrl = ""; rmClient.registerApplicationMaster(appHostName, appHostPort, appTrackingUrl); LOG.info("Register Application Master 1"); // Priority for worker containers - priorities are intra-application Priority priorityRecord = Records.newRecord(Priority.class); priorityRecord.setPriority(priority); // Resource requirements for worker containers Resource capability = Records.newRecord(Resource.class); capability.setMemory(memory); capability.setVirtualCores(vCores); // Make container requests to ResourceManager for (int i = 0; i < numberOfContainers; ++i) { ContainerRequest containerAsk = new ContainerRequest(capability, null, null, priorityRecord); LOG.info("Making resource request for [" + i + "]"); rmClient.addContainerRequest(containerAsk); } // Obtain allocated containers, launch and check for responses int responseId = 0; int completedContainers = 0; long lastReport = 0; List<Container> containers = new ArrayList<Container>(); while (completedContainers < numberOfContainers) { if (lastReport + TimeUnit.SECONDS.toNanos(5) < System.nanoTime()) { for (Container container : containers) { ContainerId containerId = container.getId(); NodeId nodeId = container.getNodeId(); ContainerStatus containerStatus = nmClient.getContainerStatus(containerId, nodeId); LOG.info("NodeId [" + nodeId + "] Container Status [" + containerStatus + "]"); // Figure out } lastReport = System.nanoTime(); } AllocateResponse response = rmClient.allocate(responseId++); for (Container container : response.getAllocatedContainers()) { containers.add(container); // Launch container by create ContainerLaunchContext ContainerLaunchContext ctx = Records.newRecord(ContainerLaunchContext.class); ctx.setCommands( Collections.singletonList(command + " 1>" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/stdout2" + " 2>" + ApplicationConstants.LOG_DIR_EXPANSION_VAR + "/stderr")); LOG.info("Launching container " + container.getId()); nmClient.startContainer(container, ctx); } for (ContainerStatus status : response.getCompletedContainersStatuses()) { completedContainers++; LOG.info("Completed container " + status.getContainerId()); } Thread.sleep(100); } // _server.stop(0); // Un-register with ResourceManager rmClient.unregisterApplicationMaster(FinalApplicationStatus.SUCCEEDED, "", ""); }